separate-stems
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides instructions for the agent to build and run shell commands using the
demucscommand-line interface to process audio files.- [EXTERNAL_DOWNLOADS]: As noted inSKILL.md, thedemucsutility will automatically download approximately 2GB of pre-trained model weights from Meta's servers during the first run.- [INDIRECT_PROMPT_INJECTION]: - Ingestion points: The agent interpolates user-supplied values for
<input-file>,<output-dir>,<model>, and<device>directly into a shell command template inSKILL.md. - Boundary markers: There are no instructions for using delimiters or boundary markers to distinguish user-provided strings from the command structure.
- Capability inventory: The skill utilizes shell command execution via the
demucsbinary. - Sanitization: The skill lacks explicit sanitization or escaping instructions for user-provided inputs, which could allow for command injection if a user provides a filename containing shell metacharacters (e.g.,
; rm -rf /).
Audit Metadata